基于目标区域的卷积神经网络火灾烟雾识别 
下载： 1086次
Convolutional Neural Network Fire Smoke Detection Based on Target Region
西安建筑科技大学信息与控制工程学院, 陕西 西安 710055
图 & 表
图 1. 背景差分法流程图
Fig. 1. Flowchart of background difference method
下载图片 查看原文
图 2. MD_CNN模型结构图
Fig. 2. MD_CNN model structure diagram
下载图片 查看原文
图 3. MD_CNN模型识别火灾烟雾流程图
Fig. 3. Flow chart of MD_CNN model detection fire smoke
下载图片 查看原文
图 4. 烟雾目标区域。(a)原图;(b)烟雾阈值图像;(c)烟雾目标区域
Fig. 4. Smoke target region. (a) Original image; (b) smoke threshold image; (c) smoke target region
下载图片 查看原文
图 5. ReLU激活函数图
Fig. 5. ReLU activation function
下载图片 查看原文
表 1烟雾识别层参数
Table1. Smoke detection layer parameters
| Type | Feature map size | Kernel size | Stride |
|---|
| Conv1 | 320×240×32 | 5×5 | 1 | | Pool1 | 160×120×32 | 2×2 | 2 | | Conv2 | 160×120×64 | 5×5 | 1 | | Pool2 | 80×60×64 | 2×2 | 2 | | Conv3 | 80×60×128 | 3×3 | 1 | | Pool3 | 40×30×128 | 2×2 | 2 | | Conv4 | 40×30×128 | 3×3 | 1 | | Pool4 | 20×15×128 | 2×2 | 2 | | FC5 | 1024 | | | | FC6 | 1024 | | | | FC7 | 2 | | |
|
查看原文
表 2CNN和MD_CNN的烟雾识别结果
Table2. Smoke recognition results of CNN and MD_CNN%
| Algorithm | ACC | TPR | FNR |
|---|
| CNN | 89.57 | 94.28 | 10.43 | | MD_CNN | 93.72 | 97.25 | 6.28 |
|
查看原文
表 3不同模型的结构及性能指标对比
Table3. Comparison of structure and performance indicators of different models
| Model | ACC /% | TPR /% | FNR /% | Number of Layers |
|---|
| Conv | Pooling | FC |
|---|
| Conv2_FC3 | 83.52 | 89.73 | 16.48 | 2 | 2 | 3 | | Conv3_FC3 | 88.43 | 93.68 | 11.57 | 3 | 3 | 3 | | Conv4_FC3 | 93.72 | 97.25 | 6.28 | 4 | 4 | 3 | | Conv5_FC3 | 92.93 | 95.89 | 7.07 | 5 | 5 | 3 | | Conv6_FC3 | 92.16 | 96.54 | 7.84 | 6 | 6 | 3 | | Conv7_FC3 | 90.04 | 94.39 | 9.96 | 7 | 7 | 3 |
|
查看原文
表 4四种算法的分类性能指标值
Table4. Classification performance index values of four algorithms%
| Algorithm | ACC | TPR | FNR |
|---|
| MD_CNN | 93.72 | 97.25 | 6.28 | | Algorithm 1 | 83.27 | 94.13 | 11.24 | | Algorithm 2 | 86.53 | 95.62 | 9.13 | | Algorithm 3 | 89.75 | 96.32 | 8.54 |
|
查看原文
冯路佳, 王慧琴, 王可, 卢英, 王钾. 基于目标区域的卷积神经网络火灾烟雾识别[J]. 激光与光电子学进展, 2020, 57(16): 161004. Lujia Feng, Huiqin Wang, Ke Wang, Ying Lu, Jia Wang. Convolutional Neural Network Fire Smoke Detection Based on Target Region[J]. Laser & Optoelectronics Progress, 2020, 57(16): 161004.
PDF全文
